V-Nova stepped out of stealth this week in the middle of London, UK, amazed that it had kept its own secret for the past five years, with the claim that it can revolutionize video codecs. Working under the banner of – Video as nature intended, and referring to how the human brain copes with processing video in just a few kilobytes of bandwidth – the company says that it will insert itself into every layer of video processing and save broadcasters and operators countless $billions in capex.
There was much hoopla and marketing speak at the launch of the new V-Nova technology this week, but believe the company’s claims or not, given that this is the fourth announcement that Faultline has covered this year which intends to cut in half video data rates – one thing is clear – genuine innovators are back in fashion, looking at algorithms for the way ahead to save $billions and $billions in network build out costs, as cellular operators in particular flounder over challenges on the near horizon.
The problem has been staring at us for the past five years, and didn’t need the oft quoted Cisco VNI to point out – simply put, a consumer expectation has emerged, for a variety of reasons, that cheap, high quality video should be part of the mobile experience and anyone who can deliver it at a profit, go the spoils.
V-Nova may well be one of the start-ups that contribute to solving a problem which Qualcomm analyzed correctly first – to offer 1,000 times the data rates over a five year period. The difference between the Qualcomm answer to that problem and the answers provided by companies like V-Nova, and we would add that Magnacom, Beamr and Faroudja should all be put into this bracket – is that Qualcomm rubs its hands at massive capex inflation at the heart of the cellular network, while these algorithmic approaches simply see a new arrangement of what is already there.
We think it is time that the cellular equipment vendors, Qualcomm in particular, were brought kicking and screaming into this “thinking outside the box” approach that these smaller companies have gone in for. Qualcomm started life on the back of two inventions, a revolution in forward error correction and CDMA, a brand new way of modulating radio waves – today it has become a chip machine which can bring difficult cellular chips to market 9 months ahead of the pack.
Here are the bald claims of V-Nova in a sentence – HD video on a phone at 320Kbps with no new equipment, just new software – but until we get a deeper explanation of how it works, and further endorsements from operators, we cannot speak for these claims.
The company says that it has simply rethought the way codecs work and has come out with a new approach it calls Perseus, named after the Greek demigod who overcame the Gorgons, including Medusa, and defied the Gods, Poseidon in particular, and through bold deeds won the favor of his father Zeus. So think of him as a bit of a bad boy.
Guido Meardi CEO of V-Nova is convincing. He told us that video was hierarchical, and that when we first look at someone we look at the core facial details first and then draw back and take in their hair, their suit and that this was supposed to give us some confidence in the company’s new approach.
Further clues came throughout the session – “Today’s Codecs like HEVC, are not taking advantage of the chip power that is available to them, they were built around chips designed in the 70s, 80s, and 90s, but today chips offer huge amounts of parallel processing.”
“Each codec is designed for a single output – if you want HD, that is one pass through the codec and you have to re-encode for every other resolution. V-Nova has a single output, and which can be used for any resolution. It focuses on understanding the correlation between space and time and relies heavily on prediction statistics.
“In most codecs prediction has become less and less important. We rely on them a lot, but there is no single trick. You can’t use frequency transforms, or wavelets to do this, they don’t scale – you need more prediction.” Interestingly the first thing that mobile phone codecs throw out are the complex mathematical Cabac transforms, because although they allow greater compression through prediction, they use a lot of battery power. Meardi said, “That’s true, but our predictions take up a lot less processing, so it’s not a problem. In fact our system uses less battery power than other codecs, not more.
Meardi through Q&A and general discussion revealed more technical details. The system does not rely on I-Frames, but in order to co-exist with existing infrastructure, it can work with existing encoders and use I-Frames (we assume there some pre-processing required prior to the encoder).
The stream that is then created has a Perseus native version, and specific resolutions from the encoder, sent side by side. But in its native format, it has no I-frames and yet it can still be sent through adaptive bit rate protocols, broken into chunks, and the company has tried as hard as it can to work within all existing infrastructure – again we get the impression that this is more about having no disruption to making a positive decision in V-Nova’s favor, than best practice. V-Nova clearly gets its highest bit rate savings when it is run native.
Color is coded in 16 bits natively in Perseus, not quite high enough for UHD2, but high enough for UHD and to accept High Dynamic Range encoding (eventually also requiring 20 bits, but the existing commercial versions require just 10 bit).
There is definitely a case for taking a new look at the underlying math of codecs in the light of higher resolutions, bit rates and multicore architectures. HEVC merely squeezed more out of the old approach with larger and more flexible macroblocks.
The idea appears to be having continuous stream of correlated frames where you can turn up or down the degree of compression and don’t have to store or transmit multiple resolutions. In that world the idea of a mezzanine format at a set resolution goes away.
So what we assume is being sent are correlations between the pixels in subsequent and previous frames, and these are being compressed into clusters, all those which have the same pixel from prior frames all those that change by the same or similar amount along the same color plane and more of the same. Or maybe it doesn’t work at the pixel level? But we’re not sure how it would work if it didn’t.
There are all kinds of issues here. If there are no I-Frames there must be a starter frame surely, otherwise the correlations can’t be expressed. And that could lead to slower video start up. What happens in a channel change on TV? Some kind of starter frame will have to be sent or a delay will occur while it downloads absolute pixel values prior to taking advantage of correlations, and that will mean a very slow channel change. To be honest ABR channel change is really slow, so that’s not such a problem.
Some things Perseus is not. It is not natural, as it claims. That claim must go to fractal compression which used Chaos theory and reduced video to naturalist formulas. It is not a case of perceptual filtering this is what Faroudja and Beamr do – the company said that it may add this later, and that this would be additive to what the codec brings.
And it is not a standard – although it could become one – when asked directly Eric Achtmann, Executive Chairman and co-Founder said, “We’d have to consider that very carefully.”
But absent a detailed understanding of how Perseus works and without a detailed look at its patents (yes they have some and we will try to address that next week) why would we come away from such an event faintly positive about the company’s prospects?
Well the proof is in the eating and the delectation of this particular morsel was consumed by no less than DTH broadcast company Sky. V-Nova rolled out Massimo Bertolotti, head of engineering at Sky Italia. Bertolotti told us that he had at first experimented with Perseus configured as a Contribution encoder. Typically he uses JPEG 2000 (J2K) for this and sends a 140 Mbps stream from Italy’s soccer stadiums to the central live soccer editing suite at Sky Italia – Perseus offered the same quality in 75 Mbps, he said.
Bob Plumridge Chief Technology Officer in EMEA for Hitachi Data Systems also turned out at the event as the company who’s servers this ran on. We guess Hitachi plans to bring this to market to drive acceptance of its servers in cloud infrastructure. It may want to replace all the real time encoding specialists out there – Ericsson, Harmonic, Envivio, Elemental – using this system. But Plumridge instead gave as much time to speaking about health care and CCTV as applications, alongside broadcasting – and our impression was that he was not quite confident enough to chase pay TV high resolution quality video, and is likely to start down at the internet of things, sending video from your home CCTV to the cloud, a system which doesn’t need quite so much resilience.
And almost to complete the process of damning with faint praise Intel turned out to the event in the guise of its IoT manager, not someone from their I7 server division.
And given that the company claims this is supported on Broadcom chips, where was the Broadcom presence? At least it contributed a quote for the press release, “Broadcom supports the Perseus compression standard, which offers an advanced tool to effect improvements in bitrate, quality and latency. Combined with Broadcom’s set-top box technology, Perseus compression can help enable more efficient HD transmissions and more widespread Ultra HD content to subscribers around the world,” and this was attributed to Rich Nelson, Senior Vice President, Marketing, Broadband & Connectivity Group at Broadcom. That’s a quote which sounds like it went through three lawyers, two committees and a period of counselling before it emerged – sits Broadcom firmly on the fence.
There were further quotes from the European Broadcasting Union and from Tata Sky and from higher ups at Intel.
But V-Nova made it clear that there were 20 prominent members behind its consortium – but a number of them at present cannot be named. That’s either because they don’t want to be associated with Perseus until it is already a success, or because V-Nova is planning further events announcing their involvement.
The overwhelming first day news coverage has been positive – but that’s because the event was convincing, without being conclusive. But we have to ask why would supports risk damaging the chances of V-Nova by not coming out with the full set of consortium names in support.
One backer is consulting player McKinsey – because that’s where Meardi and Achtmann hail from. They met when Meardi was working at Intel on codecs and while Achtmann was creating the product development practice at McKinsey. And much of the things that were wrong with this launch were McKinsey-esque – the over expensive and unnecessary video it made aimed at the layman, not the technologist, the glossy takeaways, the name (it’s about video and it will go Nova) and the expensive central London HQ.
So does that mean that McKinsey has paid for this? Meardi let slip that he has put money in, but there is no VC funding in V-Nova, instead it has come from “broadcasters,” and in our view these have the most to gain from this process. But it may also mean that broadcasters will be making strategy decisions for V-Nova.
Right now UHD is a bridge too far, even with HEVC encoding, for broadcasters to retain their traditional route to market over the air. But if they abandon FTA delivery and opt for OTT only delivery, they risk being just another source of internet content – why would a consumer turn to a linear channel, when YouTube or Netflix has the same content for viewing anytime?
So this type of technology is seriously attractive to broadcasters, which is perhaps why Sky has got involved. Tata Sky could also benefit from this technology, broadcasting from a satellite a single signal that can drive TVs, but could also double as a mobile signal if stored at the set top.
The way Meardi describes Perseus, there is one superset of a signal, and built into that are ways of playing it out in different resolutions. Look at how say ISDB-T, the broadcasting technology in Japan has 13 segments, and offers 3 resolutions. These are discreet segments, you take one or the other, and get a single version – either HD, SD or QVGA for whichever device you use. The same is true of ATSC and ATSC M/H, one signal with two containers. But Perseus appears to be multiple layers of detail, and you can take the level you need, but we don’t think that’s how the business model would work.
Of course mobile phones in India do not have a way of receiving a satellite signal unless it is re-broadcast over WiFi, and there is little WiFi in India. The logical route to get to anything is through the cellular networks which are the most effective infrastructure on the sub-continent. But data rates on a pre-4G network are low, so a parallel delivery of unicast video over cellular for Tata Sky might give it a decisive lead. The network could certainly manage 320 Kbps but rivals without this technology would be unable to respond.
So let’s assume for the moment that this system works and that a detailed explanation, once forthcoming, stacks up.
First of all this will upset all encoder businesses, who will either be forced to license this technology or to compete with it. And if they license it, they will have to find differentiation elsewhere.
Secondly the entire HEVC community having invested in HEVC will not want to find that it’s all been in vain – and despite V-Nova’s optimism, people will remember that resentment. Google alone with its VP9 encoder will have to decide how it feels about this technology – does it want to crush it, buy it or try to ignore it?
And what would happen if this technology is acquired? If Apple bought it and it had better video compression, it would sue anyone else that copied it and this would become a force for a global an Apple monopoly. But if Apple came knocking with $1 billion, t might be hard to resist.
Meardi talked to Faultline privately about licensing costs. “It will not be more expensive than MPEG patents and we have a vested interest for it to be even cheaper.”
This is straightforward – the MPEG patents are shared among 40 or so major multinationals – that amounts to a few hundred million dollars a year. If that was just given to just a single start up, then that start up would be worth a few $ billion. That would almost certainly lead to large companies contesting the V-Nova patents, tying up the technology for five years in courts.
What V-Nova has to do is three-fold. On the one hand it needs to get this invention into a product where the consumer is aware of it – so they can express their preference. If it gets a single design win that has user uptake – that will lead to more instances and slowly an avalanche of demand. Using it for contribution encoding is not enough.
Secondly it must be used to extend the reach of effective ABR services to lower data rates, and to allow IPTV a greater reach, but here the consumer will not be aware of what was making this possible, only the operators. It could be used in broadcasting to deliver UHD into a multiplex, and that will create broadcaster UHD channels far earlier than anyone expected.
But the third and we think major use and something that would create an explosive valuation is to get this into the mobile eco-system and there are a few ways this could be done. Unencrypted Adaptive Bit Rate video could be transcoded at various points in the cellular network, at the edge, in the middle or even in a cloud RAN base station – and the last mile radio delivery would cost almost nothing in terms of bandwidth. You would take YouTube over LTE from 5 Mbps (ABR expands to take all the bandwidth it can in LTE) down to 320 kbps – and that’s a factor of ten improvement. You would need new software, built into the device, supporting all applications.
If an operator lets video expand at 100% per annum, then it still has 5 years of growth built to get back to its current level of network performance. Capex can be slowed down or even stopped for a while.
That would offset several hundreds of $billions of capex, and allow cellular operators to continue without broadcasting anything, except for the most popular live events. Any operator that uses this technology first, can price its service to fund savage market share improvements.
If you were an operator, you would feel inclined to buy the business and not let anyone else enjoy the benefit and beat them in the market. But if this came to an auction it would not go to an operator.
It’s too early right now to foretell the outcome for a technology like this. The system comes in server software, client software and a future planned silicon version that will not be available until some way down the track. But if Perseus proves itself and shows it can go onwards from where it is today, then we can see the company being hunted down, purely to create an exclusive relationship, and a market advantage.
In the end we were impressed that Sky Italia would rely on this technology for live soccer transportation, effectively its lifeblood – but the journey from there to outright success for V-Nova is a long one, fraught with obstacles – very few of which are technological.